Duckle v0.5.2

Duckle v0.5.2

Duckle v0.5.2 brings vector-database and next-generation columnar formats, a native Python UDF, query pushdown across every SQL source, new workspace settings, and a refreshed guided tour.

Dives & dashboards

Dives landed in the previous release and are front-and-center this cycle, now that the guided tour points the way to them.

• Dives - live, auto-charting data views over your pipeline output. Write a SELECT (or generate one with the AI assistant), preview it, and save it. Local-first, no server required.
• Dashboards (#100) - pin several dives onto a single page.
• Share - self-contained HTML export, plus standalone read-only /dive/<id> and /dash/<id> pages.
• Find them fast - a top-bar Dives gallery, and (new in 0.5.2) a Settings toggle to hide the button if you do not use it.

Vector & next-gen columnar formats

• LanceDB source + sink - read and write Lance datasets and tables: a local directory, LanceDB Cloud (db://), or s3:// / gs:// / az:// object stores. It runs through a small bundled sidecar so the core engine stays lean, and works out of the box with no extra install.
• Vortex source + sink (#111) - read and write the Vortex columnar file format (.vortex), a next-generation format built for fast random access. Same bundled sidecar, bridged through Parquet.
• Both show up with their official brand icons in the palette and on the canvas.

Python UDF

• code.python - a per-row transform powered by a real Python 3 interpreter, so you get the full language plus any packages you have installed. Define a process(row) function that returns the output row (return None to drop a row); rows pass in and out as JSON, so no Python runtime is baked into the engine. Point it at any interpreter with DUCKLE_PYTHON_BIN, otherwise it uses the Python on your PATH.

Query pushdown

• Custom-SQL pushdown across all relational sources - push a free-form SELECT down to the source database so filtering, joins, and aggregation run where the data lives, and only the result comes back. This is now consistent across every relational source, and exposed as a dedicated query field on BigQuery and Redshift.
• MongoDB aggregation pipelines (#106) - run a native Mongo aggregation pipeline as the source query.

Workspace settings

• Engine memory cap (#102) - set a per-workspace memory limit so a single run cannot exhaust the machine.
• Global context file - auto-load context variables from one file, so ${...} placeholders resolve from a single shared source across all your pipelines.
• Hide the Dives button - a toggle to remove the Dives icon from the top bar if you do not use it.

Onboarding & discovery

• Refreshed guided tour - now spotlights Dives (live, auto-charting data views and dashboards) and the column-lineage viewer, with broader palette coverage that calls out the 350+ components: vector databases, data-quality and governance, AI, and code UDFs (Python and JavaScript). Anyone who dismissed the earlier tour sees the new spotlights on next launch, so the flagship features stop getting missed.

Self-host: one-click Coolify deploy (#31)

• Install the web editor on a VPS (Hetzner / Hostinger / OVH, or any Coolify box)
  in one click: add a Docker Compose resource from docker-compose.coolify.yml
  and Deploy. Coolify assigns a domain with automatic HTTPS and persists your
  pipelines + data in a managed volume.
• The web image now has desktop parity for data work: the LanceDB / Vortex
  sidecar and a Python 3 interpreter (for code.python) are bundled, alongside
  the DuckDB engine. See docs/deploy-coolify.md.

Fixes

• Properties-panel node id is no longer clipped by the collapse toggle (#102).
• The web dashboard shows pipeline names instead of internal ids (#108).
• Regex Extract supports named capture groups (#109).

---

Install

Download the build for your platform below. The desktop app bundles everything it needs, including the columnar-format sidecar; nothing else to install.